RiskTorrent: Using Portfolio Optimisation for Media Streaming

Transcription

1 RiskTorrent: Using Portfolio Optimisation for Media Streaming Raul Landa, Miguel Rio Communications and Information Systems Research Group Department of Electronic and Electrical Engineering University College London

2 Definitions Reciprocity: Peers need to upload in order to obtain download capacity j Let s call the throughput that peer uploads to peer i The throughput that obtains from as a result is defined as

3 Modeling Reciprocity The simplest model for is, simply where that peer of can be thought of the return receives from, given an investment

4 Modeling Total Download Throughput The total return (throughput) for peer is then: Thus, in this model, the total return that a peer obtains is a linear combination of the throughput that it allocates to all other nodes

5 Modeling Download Throughput Variability We treat the asset returns as random variables returns have nonzero volatility The variance of, a linear combination of random variables, is then given by where is the covariance matrix of asset returns, and is the vector of assigned uploads

6 Media Streaming: The Investment View Each possible allocation of upload bandwidth to specific peers then becomes a portfolio For media streaming, we are interested in minimising throughput variability while maintaining a given stream rate In this case, swarming protocol design becomes portfolio selection [Markowitz, 1952] and [Markowitz, 1959]

7 Media Streaming: The Investment View The objective is to minimise portfolio risk while achieving a given return and satisfying a budget constraint. Diversification helps reduce risk while maintaining returns the volatility of the portfolio is smaller than that of its components. Formally:

8 Media Streaming: The Investment View The objective is to minimise throughput variability while achieving a given stream rate and satisfying a maximum upload capacity constraint. Formally: Throughput Variability Constant Stream Rate Maximum Upload Capacity

9 Media Streaming: The Investment View Throughput Variability Constant Stream Rate Maximum Upload Capacity Non-negativity Constraints (no short-selling)

10 Media Streaming: The Investment View Throughput Variability Constant Stream Rate Maximum Upload Capacity Non-negativity Constraints (no short-selling) What happens if the problem is unfeasible?

11 Media Streaming: The Investment View Usually, this means that the peer has insufficient upload capacity (capital) to sustain the required stream rate (return) In this case, peers fall back to maximising throughput, irrespective of risk:

12 Media Streaming: The Investment View Usually, this means that the peer has insufficient upload capacity (capital) to sustain the required stream rate (return) In this case, peers fall back to maximising throughput, irrespective of risk: Stream Rate Maximum Upload Capacity Non-negativity Constraints (no short-selling)

13 Simulations: Setup (Expected Returns) A 1 B C D

14 Simulations: Setup (Covariance Matrix) A D B C

15 Simulations: Achievable Stream Rate

16 Simulations: Risk (Standard Deviation)

17 Simulations: Protocol Operation Curves

18 Conclusions A possible model for reciprocity-based peer-topeer networks can be formulated based on portfolio optimisation The model can be extended: Multi-stage formulations Asymmetric risk measures More general reciprocity models See [Steinbach, 2001] and references therein Practical issues: How can we measure the covariance matrix?

19 Thank You! Any Questions?

20 References Markowitz, H. M. (1952) Portfolio Selection. The Journal of Finance 7 (1): Markowitz, H.M. (1959) Portfolio Selection: Efficient Diversification of Investments. John Wiley & Sons. Steinbach, M. C. (2001) Markowitz Revisited: Mean-Variance Models in Financial Portfolio Analysis. SIAM Rev. 43 (1): 31-85